Models of vegetation composition based on climate predictors

Based on code from Martin Holdrege’s Sagebrush-Fire paper

The data consists of vegetation % cover by functional group from across CONUS (from AIM, FIA, LANDFIRE, and RAP), as well as climate variables from DayMet, which have been aggregated into mean interannual conditions accross multiple temporal windows.

Dependencies

User defined parameters

print(params)

## $run
## [1] TRUE
## 
## $test_run
## [1] FALSE
## 
## $save_figs
## [1] FALSE
## 
## $ecoregion
## [1] "forest"
## 
## $response
## [1] "ShrubCover"

# set to true if want to run for a limited number of rows (i.e. for code testing)
test_run <- params$test_run
save_figs <- params$save_figs
response <- params$response
fit_sample <- TRUE # fit model to a sample of the data
n_train <- 5e4 # sample size of the training data
n_test <- 1e6 # sample size of the testing data (if this is too big the decile dotplot code throws memory errors)


run <- params$run

# set option so resampled dataset created here reproduces earlier runs of this code with dplyr 1.0.10
source("../../../Functions/glmTransformsIterates.R")
source("../../../Functions/transformPreds.R")
source("../../../Functions/StepBeta_mine.R")
#source("src/fig_params.R")
#source("src/modeling_functions.R")
 
library(ggspatial)
library(terra)
library(tidyterra)
library(sf)
library(caret)
library(tidyverse)
library(GGally) # for ggpairs()
library(pdp) # for partial dependence plots
library(gridExtra)
library(knitr)
library(patchwork) # for figure insets etc. 
library(ggtext)
library(StepBeta)
theme_set(theme_classic())
library(here)
library(rsample)
library(kableExtra)
library(glmnet)
library(USA.state.boundaries)

read in data

Data compiled in the prepDataForModels.R script

here::i_am("Analysis/VegComposition/ModelFitting/02_ModelFitting.Rmd")
modDat <- readRDS( here("Data_processed", "CoverData", "DataForModels_spatiallyAveraged_withSoils_noSf.rds"))
## there are some values of the annual wet degree days 5th percentile that have -Inf?? change to lowest value for now? 
modDat[is.infinite(modDat$annWetDegDays_5percentile_3yrAnom), "annWetDegDays_5percentile_3yrAnom"] <- -47.8
## same, but for annual water deficit 95th percentile 
modDat[is.infinite(modDat$annWaterDeficit_95percentile_3yrAnom), "annWaterDeficit_95percentile_3yrAnom"] <- -600

# # Convert total cover variables into proportions (for later use in beta regression models) ... proportions are already scaled from zero to 1
# modDat <- modDat %>%
#   mutate(TotalTreeCover = TotalTreeCover/100,
#          CAMCover = CAMCover/100,
#          TotalHerbaceousCover = TotalHerbaceousCover/100,
#          BareGroundCover = BareGroundCover/100,
#          ShrubCover = ShrubCover/100
#          )
# For all response variables, make sure there are no 0s add or subtract .0001 from each, since the Gamma model framework can't handle that
modDat[modDat$TotalTreeCover == 0 & !is.na(modDat$TotalTreeCover), "TotalTreeCover"] <- 0.0001
modDat[modDat$CAMCover == 0 & !is.na(modDat$CAMCover), "CAMCover"] <- 0.0001
modDat[modDat$TotalHerbaceousCover == 0  & !is.na(modDat$TotalHerbaceousCover), "TotalHerbaceousCover"] <- 0.0001
modDat[modDat$BareGroundCover == 0 & !is.na(modDat$BareGroundCover), "BareGroundCover"] <- 0.0001
modDat[modDat$ShrubCover == 0 & !is.na(modDat$ShrubCover), "ShrubCover"] <- 0.0001
modDat[modDat$BroadleavedTreeCover_prop == 0 & !is.na(modDat$BroadleavedTreeCover_prop), "BroadleavedTreeCover_prop"] <- 0.0001
modDat[modDat$NeedleLeavedTreeCover_prop == 0 & !is.na(modDat$NeedleLeavedTreeCover_prop), "NeedleLeavedTreeCover_prop"] <- 0.0001
modDat[modDat$C4Cover_prop == 0 & !is.na(modDat$C4Cover_prop), "C4Cover_prop"] <- 0.0001
modDat[modDat$C3Cover_prop == 0 & !is.na(modDat$C3Cover_prop), "C3Cover_prop"] <- 0.0001
modDat[modDat$ForbCover_prop == 0 & !is.na(modDat$ForbCover_prop), "ForbCover_prop"] <- 0.0001
# 
# modDat[modDat$TotalTreeCover ==1& !is.na(modDat$TotalTreeCover), "TotalTreeCover"] <- 0.999
# modDat[modDat$CAMCover ==1& !is.na(modDat$CAMCover), "CAMCover"] <- 0.999
# modDat[modDat$TotalHerbaceousCover ==1 & !is.na(modDat$TotalHerbaceousCover), "TotalHerbaceousCover"] <- 0.999
# modDat[modDat$BareGroundCover ==1& !is.na(modDat$BareGroundCover), "BareGroundCover"] <- 0.999
# modDat[modDat$ShrubCover ==1& !is.na(modDat$ShrubCover), "ShrubCover"] <- 0.999
# modDat[modDat$BroadleavedTreeCover_prop ==1& !is.na(modDat$BroadleavedTreeCover_prop), "BroadleavedTreeCover_prop"] <- 0.999
# modDat[modDat$NeedleLeavedTreeCover_prop ==1& !is.na(modDat$NeedleLeavedTreeCover_prop), "NeedleLeavedTreeCover_prop"] <- 0.999
# modDat[modDat$C4Cover_prop ==1& !is.na(modDat$C4Cover_prop), "C4Cover_prop"] <- 0.999
# modDat[modDat$C3Cover_prop ==1& !is.na(modDat$C3Cover_prop), "C3Cover_prop"] <- 0.999
# modDat[modDat$ForbCover_prop ==1& !is.na(modDat$ForbCover_prop), "ForbCover_prop"] <- 0.999

Prep data

set.seed(1234)
modDat_1 <- modDat %>% 
  select(-c(prcp_annTotal:annVPD_min)) %>% 
  # mutate(Lon = st_coordinates(.)[,1], 
  #        Lat = st_coordinates(.)[,2])  %>% 
  # st_drop_geometry() %>% 
  # filter(!is.na(newRegion))
  rename("tmin" = tmin_meanAnnAvg_CLIM, 
     "tmax" = tmax_meanAnnAvg_CLIM, #1
     "tmean" = tmean_meanAnnAvg_CLIM, 
     "prcp" = prcp_meanAnnTotal_CLIM, 
     "t_warm" = T_warmestMonth_meanAnnAvg_CLIM,
     "t_cold" = T_coldestMonth_meanAnnAvg_CLIM, 
     "prcp_wet" = precip_wettestMonth_meanAnnAvg_CLIM,
     "prcp_dry" = precip_driestMonth_meanAnnAvg_CLIM, 
     "prcp_seasonality" = precip_Seasonality_meanAnnAvg_CLIM, #2
     "prcpTempCorr" = PrecipTempCorr_meanAnnAvg_CLIM,  #3
     "abvFreezingMonth" = aboveFreezing_month_meanAnnAvg_CLIM, 
     "isothermality" = isothermality_meanAnnAvg_CLIM, #4
     "annWatDef" = annWaterDeficit_meanAnnAvg_CLIM, 
     "annWetDegDays" = annWetDegDays_meanAnnAvg_CLIM,
     "VPD_mean" = annVPD_mean_meanAnnAvg_CLIM, 
     "VPD_max" = annVPD_max_meanAnnAvg_CLIM, #5
     "VPD_min" = annVPD_min_meanAnnAvg_CLIM, #6
     "VPD_max_95" = annVPD_max_95percentile_CLIM, 
     "annWatDef_95" = annWaterDeficit_95percentile_CLIM, 
     "annWetDegDays_5" = annWetDegDays_5percentile_CLIM, 
     "frostFreeDays_5" = durationFrostFreeDays_5percentile_CLIM, 
     "frostFreeDays" = durationFrostFreeDays_meanAnnAvg_CLIM, 
     "soilDepth" = soilDepth, #7
     "clay" = surfaceClay_perc, 
     "sand" = avgSandPerc_acrossDepth, #8
     "coarse" = avgCoarsePerc_acrossDepth, #9
     "carbon" = avgOrganicCarbonPerc_0_3cm, #10
     "AWHC" = totalAvailableWaterHoldingCapacity,
     ## anomaly variables
     tmean_anom = tmean_meanAnnAvg_3yrAnom, #15
     tmin_anom = tmin_meanAnnAvg_3yrAnom, #16
     tmax_anom = tmax_meanAnnAvg_3yrAnom, #17
    prcp_anom = prcp_meanAnnTotal_3yrAnom, #18
      t_warm_anom = T_warmestMonth_meanAnnAvg_3yrAnom,  #19
     t_cold_anom = T_coldestMonth_meanAnnAvg_3yrAnom, #20
      prcp_wet_anom = precip_wettestMonth_meanAnnAvg_3yrAnom, #21
      precp_dry_anom = precip_driestMonth_meanAnnAvg_3yrAnom,  #22
    prcp_seasonality_anom = precip_Seasonality_meanAnnAvg_3yrAnom, #23 
     prcpTempCorr_anom = PrecipTempCorr_meanAnnAvg_3yrAnom, #24
      aboveFreezingMonth_anom = aboveFreezing_month_meanAnnAvg_3yrAnom, #25  
    isothermality_anom = isothermality_meanAnnAvg_3yrAnom, #26
       annWatDef_anom = annWaterDeficit_meanAnnAvg_3yrAnom, #27
     annWetDegDays_anom = annWetDegDays_meanAnnAvg_3yrAnom,  #28
      VPD_mean_anom = annVPD_mean_meanAnnAvg_3yrAnom, #29
      VPD_min_anom = annVPD_min_meanAnnAvg_3yrAnom,  #30
      VPD_max_anom = annVPD_max_meanAnnAvg_3yrAnom,  #31
     VPD_max_95_anom = annVPD_max_95percentile_3yrAnom, #32
      annWatDef_95_anom = annWaterDeficit_95percentile_3yrAnom, #33 
      annWetDegDays_5_anom = annWetDegDays_5percentile_3yrAnom ,  #34
    frostFreeDays_5_anom = durationFrostFreeDays_5percentile_3yrAnom, #35 
      frostFreeDays_anom = durationFrostFreeDays_meanAnnAvg_3yrAnom #36
  )

# small dataset for if testing the data
if(test_run) {
  modDat_1 <- slice_sample(modDat_1, n = 1e5)
}

Add a constant to the response variable (+1) so that models run…

modDat_1[,response] <- modDat_1[,response]+2

Identify the ecoregion and response variable type to use in this model run

ecoregion <- params$ecoregion
response <- params$response
print(paste0("In this model run, the ecoregion is ", ecoregion," and the response variable is ",response))

## [1] "In this model run, the ecoregion is forest and the response variable is ShrubCover"

Subset the data to only include data for the ecoregion of interest

if (ecoregion == "shrubGrass") {
  # select data for the ecoregion of interest
  modDat_1 <- modDat_1 %>%
    filter(newRegion == "dryShrubGrass")
} else if (ecoregion == "forest") {
  # select data for the ecoregion of interest
  modDat_1 <- modDat_1 %>% 
    filter(newRegion %in% c("eastForest", "westForest"))
}

# remove the rows that have no observations for the response variable of interest
modDat_1 <- modDat_1[!is.na(modDat_1[,response]),]

Currently, subsampling data from the “Texas Coastal Plain”, since it’s quite different from other regions and is really messing with model fit

modDat_1_noLA <- modDat_1 %>% 
  filter(NA_L2NAME != "TEXAS-LOUISIANA COASTAL PLAIN")
modDat_1_LA <- modDat_1 %>% 
  filter(NA_L2NAME == "TEXAS-LOUISIANA COASTAL PLAIN")
# sample points 
modDat_1 <- modDat_1_LA %>% 
  slice_sample(n = round(nrow(modDat_1_LA)*.3)) %>% 
  rbind(modDat_1_noLA) 

Visualize the response variable

hist(modDat_1[,response], main = paste0("Histogram of ",response),
     xlab = paste0(response))

Visualize the predictor variables

The following are the candidate predictor variables for this ecoregion:

if (ecoregion == "shrubGrass") {
  # select potential predictor variables for the ecoregion of interest
        prednames <-
          c(
"tmean"             , "prcp"                    ,"prcp_seasonality"        ,"prcpTempCorr"          , 
"isothermality"     , "annWatDef"               ,"sand"                    ,"coarse"                , 
"carbon"            , "AWHC"                    ,"tmin_anom"               ,"tmax_anom"             , 
"t_warm_anom"       , "prcp_wet_anom"           ,"precp_dry_anom"          ,"prcp_seasonality_anom" , 
"prcpTempCorr_anom" , "aboveFreezingMonth_anom" ,"isothermality_anom"      ,"annWatDef_anom"        , 
"annWetDegDays_anom", "VPD_mean_anom"           ,"VPD_min_anom"            ,"frostFreeDays_5_anom"   )
  
} else if (ecoregion == "forest") {
  # select potential predictor variables for the ecoregion of interest
  prednames <- 
    c(
"tmean"                 ,"prcp"               , "prcp_dry"                , "prcpTempCorr"      ,     
"isothermality"         ,"annWatDef"          , "clay"                    , "sand"              ,     
"coarse"                ,"carbon"             , "AWHC"                    , "tmin_anom"         ,     
"tmax_anom"             ,"prcp_anom"          , "prcp_wet_anom"           , "precp_dry_anom"    ,     
"prcp_seasonality_anom" ,"prcpTempCorr_anom"  , "aboveFreezingMonth_anom" , "isothermality_anom",     
"annWatDef_anom"        ,"annWetDegDays_anom" , "VPD_mean_anom"           , "VPD_max_95_anom"   ,     
"frostFreeDays_5_anom"   )
}

# subset the data to only include these predictors, and remove any remaining NAs 
modDat_1 <- modDat_1 %>% 
  select(prednames, response, newRegion, Year, Long, Lat, NA_L1NAME, NA_L2NAME) %>% 
  drop_na()

names(prednames) <- prednames
df_pred <- modDat_1[, prednames]
# 
# # print the list of predictor variables
# knitr::kable(format = "html", data.frame("Possible_Predictors" = prednames)
# ) %>%
#   kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

create_summary <- function(df) {
  df %>% 
    pivot_longer(cols = everything(),
                 names_to = 'variable') %>% 
    group_by(variable) %>% 
    summarise(across(value, .fns = list(mean = ~mean(.x, na.rm = TRUE), min = ~min(.x, na.rm = TRUE), 
                                        median = ~median(.x, na.rm = TRUE), max = ~max(.x, na.rm = TRUE)))) %>% 
    mutate(across(where(is.numeric), round, 4))
}

modDat_1[prednames] %>% 
  create_summary() %>% 
  knitr::kable(caption = 'summaries of possible predictor variables') %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

summaries of possible predictor variables

variable	value_mean	value_min	value_median	value_max
AWHC	14.4926	0.9085	13.6502	34.6680
VPD_max_95_anom	0.0597	-0.5726	0.0380	0.8326
VPD_mean_anom	-0.0218	-0.3148	-0.0216	0.2651
aboveFreezingMonth_anom	0.1132	-1.8939	0.0667	3.3667
annWatDef	34.9209	0.0000	25.0013	303.5579
annWatDef_anom	-0.0563	-8.9984	-0.0758	1.0000
annWetDegDays_anom	-0.0060	-1.2398	-0.0204	0.8050
carbon	7.8926	0.2171	4.4906	51.0604
clay	16.1406	0.0286	15.9896	83.0975
coarse	15.2101	0.0000	13.4007	64.1122
frostFreeDays_5_anom	-25.6654	-273.1000	-30.0000	53.1000
isothermality	36.2187	21.4503	36.4480	59.8804
isothermality_anom	0.7373	-8.4018	0.7307	11.7898
prcp	1103.4888	210.3373	1125.1102	4095.4507
prcpTempCorr	-0.1559	-0.8596	-0.1240	0.7258
prcpTempCorr_anom	-0.0121	-0.5978	-0.0003	0.6098
prcp_anom	0.0044	-0.8885	0.0011	0.6720
prcp_dry	16.8895	0.0003	12.5327	76.7440
prcp_seasonality_anom	-0.0042	-0.6092	0.0032	0.4788
prcp_wet_anom	0.0035	-1.4179	0.0130	0.6981
precp_dry_anom	0.0658	-9.0000	0.0769	1.0000
sand	45.7520	0.0098	45.2042	99.9418
tmax_anom	-0.2825	-5.6017	-0.2889	4.0197
tmean	9.4406	-2.4480	8.2771	24.9713
tmin_anom	-0.6272	-5.7692	-0.5726	2.6920

# response_summary <- modDat_1 %>% 
#     dplyr::select(#where(is.numeric), -all_of(pred_vars),
#       matches(response)) %>% 
#     create_summary()
# 
# 
# kable(response_summary, 
#       caption = 'summaries of response variables, calculated using paint') %>%
# kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

Plot predictor vars against each other

set.seed(12011993)
# function for colors
my_fn <- function(data, mapping, method="p", use="pairwise", ...){
  
  # grab data
  x <- eval_data_col(data, mapping$x)
  y <- eval_data_col(data, mapping$y)
  
  # calculate correlation
  corr <- cor(x, y, method=method, use=use)
  
  # calculate colour based on correlation value
  # Here I have set a correlation of minus one to blue, 
  # zero to white, and one to red 
  # Change this to suit: possibly extend to add as an argument of `my_fn`
  colFn <- colorRampPalette(c("red", "white", "blue"), interpolate ='spline')
  fill <- colFn(100)[findInterval(corr, seq(-1, 1, length=100))]
  
  ggally_cor(data = data, mapping = mapping, size = 2.5, stars = FALSE, 
             digits = 2, colour = I("black"),...) + 
    theme_void() +
    theme(panel.background = element_rect(fill=fill))
  
}

if (run == TRUE) {
(corrPlot <- modDat_1 %>% 
  select(prednames) %>% 
  slice_sample(n = 5e4) %>% 
  #select(-matches("_")) %>% 
ggpairs( upper = list(continuous = my_fn, size = .1), lower = list(continuous = GGally::wrap("points", alpha = 0.1, size=0.1)), progress = FALSE))
    base::saveRDS(corrPlot, paste0("../ModelFitting/models/", response, "_",ecoregion, "_corrPlot.rds"))
  
  } else {
    # corrPlot <- readRDS(paste0("../ModelFitting/models/", response, "_",ecoregion, "_corrPlot.rds"))
    # (corrPlot)
    print(c("See previous correlation figures"))
  }

Predictor variables compared to binned response variables

set.seed(12011993)
# vector of name of response variables
vars_response <- response

# longformat dataframes for making boxplots
df_sample_plots <-  modDat_1  %>% 
  slice_sample(n = 5e4) %>% 
   rename(response = all_of(response)) %>% 
  mutate(response = case_when(
    response <= .25 ~ ".25", 
    response > .25 & response <=.5 ~ ".5", 
    response > .5 & response <=.75 ~ ".75", 
    response >= .75  ~ "1", 
  )) %>% 
  select(c(response, prednames)) %>% 
  tidyr::pivot_longer(cols = unname(prednames), 
               names_to = "predictor", 
               values_to = "value"
               )  
 

  ggplot(df_sample_plots, aes_string(x= "response", y = 'value')) +
  geom_boxplot() +
  facet_wrap(~predictor , scales = 'free_y') + 
  ylab("Predictor Variable Values") + 
    xlab(response)

Standardize the predictor variables for the model-fitting process

modDat_1_s <- modDat_1 %>% 
  mutate(across(all_of(prednames), base::scale, .names = "{.col}_s")) 
names(modDat_1_s) <- c(names(modDat_1),
                       paste0(prednames, "_s")
                       )
  
scaleFigDat_1 <- modDat_1_s %>% 
  select(c(Long, Lat, Year, prednames)) %>% 
  pivot_longer(cols = all_of(names(prednames)), 
               names_to = "predNames", 
               values_to = "predValues_unScaled")
scaleFigDat_2 <- modDat_1_s %>% 
  select(c(Long, Lat, Year,paste0(prednames, "_s"))) %>% 
  pivot_longer(cols = all_of(paste0(prednames, "_s")), 
               names_to = "predNames", 
               values_to = "predValues_scaled", 
               names_sep = ) %>% 
  mutate(predNames = str_replace(predNames, pattern = "_s$", replacement = ""))

scaleFigDat_3 <- scaleFigDat_1 %>% 
  left_join(scaleFigDat_2)

ggplot(scaleFigDat_3) + 
  facet_wrap(~predNames, scales = "free") +
  geom_histogram(aes(predValues_unScaled), fill = "lightgrey", col = "darkgrey") + 
  geom_histogram(aes(predValues_scaled), fill = "lightblue", col = "blue") +
  xlab ("predictor variable values") + 
  ggtitle("Comparing the distribution of unscaled (grey) to scaled (blue) predictor variables")

Model Fitting

Visualize the level 2 ecoregions and how they differ across environmental space

## visualize the variation between groups across environmental space

## do a pca of climate across level 2 ecoregions
pca <- prcomp(modDat_1_s[,paste0(prednames, "_s")])
library(factoextra)
(fviz_pca_ind(pca, habillage = modDat_1_s$NA_L2NAME, label = "none", addEllipses = TRUE, ellipse.level = .95, ggtheme = theme_minimal(), alpha.ind = .1))

if (ecoregion == "shrubGrass") {
  print("We'll combine the 'Mediterranean California' and 'Western Sierra Madre Piedmont' ecoregions (into 'Mediterranean Piedmont'). We'll also combine `Tamaulipas-Texas semiarid plain,' 'Texas-Lousiana Coastal plain,' and 'South Central semiarid prairies' ecoregions (into (`Semiarid plain and prairies`)." )
  
  modDat_1_s[modDat_1_s$NA_L2NAME %in% c("MEDITERRANEAN CALIFORNIA", "WESTERN SIERRA MADRE PIEDMONT"), "NA_L2NAME"] <- "MEDITERRANEAN PIEDMONT"
  modDat_1[modDat_1$NA_L2NAME %in% c("MEDITERRANEAN CALIFORNIA", "WESTERN SIERRA MADRE PIEDMONT"), "NA_L2NAME"] <- "MEDITERRANEAN PIEDMONT"
  
  modDat_1_s[modDat_1_s$NA_L2NAME %in% c("TAMAULIPAS-TEXAS SEMIARID PLAIN", "TEXAS-LOUISIANA COASTAL PLAIN", "SOUTH CENTRAL SEMIARID PRAIRIES"), "NA_L2NAME"] <- "SEMIARID PLAIN AND PRAIRIES"
  modDat_1[modDat_1$NA_L2NAME %in% c("TAMAULIPAS-TEXAS SEMIARID PLAIN", "TEXAS-LOUISIANA COASTAL PLAIN", "SOUTH CENTRAL SEMIARID PRAIRIES"), "NA_L2NAME"] <- "SEMIARID PLAIN AND PRAIRIES"
}
# make a table of n for each region
modDat_1 %>% 
  group_by(NA_L2NAME) %>% 
  dplyr::summarize("Number_Of_Observations" = length(NA_L2NAME)) %>% 
  rename("Level_2_Ecoregion" = NA_L2NAME)%>% 
  kable() %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

Level_2_Ecoregion	Number_Of_Observations
ATLANTIC HIGHLANDS	5598
CENTRAL USA PLAINS	1253
EVERGLADES	211
MARINE WEST COAST FOREST	7402
MISSISSIPPI ALLUVIAL AND SOUTHEAST USA COASTAL PLAINS	9718
MIXED WOOD PLAINS	8087
MIXED WOOD SHIELD	7930
OZARK/OUACHITA-APPALACHIAN FORESTS	15472
SOUTHEASTERN USA PLAINS	23479
UPPER GILA MOUNTAINS	8759
WESTERN CORDILLERA	71139

Then, look at the spatial distribution and environmental characteristics of the grouped ecoregions

## make data into spatial format
modDat_1_sf <- modDat_1 %>% 
  st_as_sf(coords = c("Long", "Lat"), crs = st_crs("PROJCRS[\"unnamed\",\n    BASEGEOGCRS[\"unknown\",\n        DATUM[\"unknown\",\n            ELLIPSOID[\"Spheroid\",6378137,298.257223563,\n                LENGTHUNIT[\"metre\",1,\n                    ID[\"EPSG\",9001]]]],\n        PRIMEM[\"Greenwich\",0,\n            ANGLEUNIT[\"degree\",0.0174532925199433,\n                ID[\"EPSG\",9122]]]],\n    CONVERSION[\"Lambert Conic Conformal (2SP)\",\n        METHOD[\"Lambert Conic Conformal (2SP)\",\n            ID[\"EPSG\",9802]],\n        PARAMETER[\"Latitude of false origin\",42.5,\n            ANGLEUNIT[\"degree\",0.0174532925199433],\n            ID[\"EPSG\",8821]],\n        PARAMETER[\"Longitude of false origin\",-100,\n            ANGLEUNIT[\"degree\",0.0174532925199433],\n            ID[\"EPSG\",8822]],\n        PARAMETER[\"Latitude of 1st standard parallel\",25,\n            ANGLEUNIT[\"degree\",0.0174532925199433],\n            ID[\"EPSG\",8823]],\n        PARAMETER[\"Latitude of 2nd standard parallel\",60,\n            ANGLEUNIT[\"degree\",0.0174532925199433],\n            ID[\"EPSG\",8824]],\n        PARAMETER[\"Easting at false origin\",0,\n            LENGTHUNIT[\"metre\",1],\n            ID[\"EPSG\",8826]],\n        PARAMETER[\"Northing at false origin\",0,\n            LENGTHUNIT[\"metre\",1],\n            ID[\"EPSG\",8827]]],\n    CS[Cartesian,2],\n        AXIS[\"easting\",east,\n            ORDER[1],\n            LENGTHUNIT[\"metre\",1,\n                ID[\"EPSG\",9001]]],\n        AXIS[\"northing\",north,\n            ORDER[2],\n            LENGTHUNIT[\"metre\",1,\n                ID[\"EPSG\",9001]]]]"))


# download map info for visualization
data(state_boundaries_wgs84) 

cropped_states <- suppressMessages(state_boundaries_wgs84 %>%
  dplyr::filter(NAME!="Hawaii") %>%
  dplyr::filter(NAME!="Alaska") %>%
  dplyr::filter(NAME!="Puerto Rico") %>%
  dplyr::filter(NAME!="American Samoa") %>%
  dplyr::filter(NAME!="Guam") %>%
  dplyr::filter(NAME!="Commonwealth of the Northern Mariana Islands") %>%
  dplyr::filter(NAME!="United States Virgin Islands") %>%

  sf::st_sf() %>%
  sf::st_transform(sf::st_crs(modDat_1_sf))) #%>%
  #sf::st_crop(sf::st_bbox(modDat_1_sf)+c(-1,-1,1,1))

map1 <- ggplot() +
  geom_sf(data=cropped_states,fill='white') +
  geom_sf(data=modDat_1_sf,aes(fill=as.factor(NA_L2NAME)),linewidth=0.5,alpha=0.5) +
  geom_point(data=modDat_1,alpha=0.5, 
             aes(x = Long, y = Lat, color=as.factor(NA_L2NAME)), alpha = .1) +
  #scale_fill_okabeito() +
  #scale_color_okabeito() +
 # theme_default() +
  theme(legend.position = 'none') +
  labs(title = "Level 2 Ecoregions as spatial blocks")

hull <- modDat_1_sf %>%
  ungroup() %>%
  group_by(NA_L2NAME) %>%
  slice(chull(tmean, prcp))

plot1<-ggplot(data=modDat_1_sf,aes(x=tmean,y=prcp)) +
  geom_polygon(data = hull, alpha = 0.25,aes(fill=NA_L2NAME) )+
  geom_point(aes(group=NA_L2NAME,color=NA_L2NAME),alpha=0.25) +
  theme_minimal() + xlab("Annual Average T_mean - long-term average") +
  ylab("Annual Average Precip - long-term average") #+
  #scale_color_okabeito() +
  #scale_fill_okabeito()

plot2<-ggplot(data=modDat_1_sf %>%
                pivot_longer(cols=tmean:prcp),
              aes(x=value,group=name)) +
  # geom_polygon(data = hull, alpha = 0.25,aes(fill=fold) )+
  geom_density(aes(group=NA_L2NAME,fill=NA_L2NAME),alpha=0.25) +
  theme_minimal() +
  facet_wrap(~name,scales='free')# +
  #scale_color_okabeito() +
  #scale_fill_okabeito()
 
library(patchwork)
(combo <- (map1+plot1)/plot2) 

Fit a global model with all of the data

First, fit a LASSO regression model using the glmnet R package

• This regression is a Gamma glm with a log link
• Use cross validation across level 2 ecoregions to tune the lambda parameter in the LASSO model
• this model is fit to using the scaled weather/climate/soils variables
• this list of possible predictors includes:
  1. main effects
  2. interactions between all soils variables
  3. interactions between climate and weather variables
  4. transformed main effects (squared, log-transformed (add a uniform integer – 20– to all variables prior to log-transformation), square root-transformed (add a uniform integer – 20– to all variables prior to log-transformation))

## 
## Call:  cv.glmnet(x = X[, 2:ncol(X)], y = y, type.measure = "mse", foldid = my_folds,      keep = TRUE, parallel = TRUE, family = stats::Gamma(link = "log"),      alpha = 1, nlambda = 100, standardize = FALSE) 
## 
## Measure: Mean-Squared Error 
## 
##     Lambda Index Measure    SE Nonzero
## min 0.0305    31   276.1 47.36      21
## 1se 0.2844     7   322.7 55.09       3

Then, fit regular glm models (Gamma glm with a log link), first using the coefficients from the ‘best’ lambda identified in the LASSO model, as then using the coefficients from the ‘1SE’ lambda identified from the LASSO (this is the value of lambda such that the cross validation error is within 1 standard error of the minimum).

## fit w/ the identified coefficients from the 'best' lambda, but using the glm function
  mat_glmnet_best <- as.matrix(bestLambda_coef)
  mat2_glmnet_best <- mat_glmnet_best[mat_glmnet_best[,1] != 0,]
  names(mat2_glmnet_best) <- rownames(mat_glmnet_best)[mat_glmnet_best[,1] != 0]

  if (length(mat2_glmnet_best) == 1) {
    f_glm_bestLambda <- as.formula(paste0(response, "~ 1"))
  } else {
  f_glm_bestLambda <- as.formula(paste0(response, " ~ ", paste0(names( mat2_glmnet_best)[2:length(names( mat2_glmnet_best))], collapse = " + ")))
  }
  
  fit_glm_bestLambda <- glm(data = modDat_1_s
                              , formula =  f_glm_bestLambda, family =  stats::Gamma(link = "log"))
  
   ## fit w/ the identified coefficients from the '1se' lambda, but using the glm function
  mat_glmnet_1se <- as.matrix(seLambda_coef)
  mat2_glmnet_1se <- mat_glmnet_1se[mat_glmnet_1se[,1] != 0,]
  names(mat2_glmnet_1se) <- rownames(mat_glmnet_1se)[mat_glmnet_1se[,1] != 0]
  if(length(mat2_glmnet_1se) == 1) {
    f_glm_1se <- as.formula(paste0(response, "~ 1"))
  } else {
  f_glm_1se <- as.formula(paste0(response, " ~ ", paste0(names( mat2_glmnet_1se)[2:length(names( mat2_glmnet_1se))], collapse = " + ")))
  }


  fit_glm_se <- glm(data = modDat_1_s, formula = f_glm_1se,
                    family =  stats::Gamma(link = "log"))

Then, we predict (on the training set) using both of these models (best lambda and 1se lambda)

  ## predict on the test data
  # lasso model predictions with the optimal lambda
  optimal_pred <- predict(fit_glm_bestLambda, newx=X[,2:ncol(X)], type = "response") - 2
  optimal_pred_1se <-  predict(fit_glm_se, newx=X[,2:ncol(X)], type = "response") - 2
    null_fit <- glm(#data = data.frame("y" = y, X[,paste0(prednames, "_s")]), 
      formula = y ~ 1, family = stats::Gamma(link = "log"))
  null_pred <- predict(null_fit, newdata = as.data.frame(X), type = "response"
                       ) - 2

  # save data
  fullModOut <- list(
    "modelObject" = fit,
    "nullModelObject" = null_fit,
    "modelPredictions" = data.frame(#ecoRegion_holdout = rep(test_eco,length(y)),
      obs=y,
                    pred_opt=optimal_pred, 
                    pred_opt_se = optimal_pred_1se,
                    pred_null=null_pred#,
                    #pred_nopenalty=nopen_pred
                    ))
  
  
# calculate correlations between null and optimal model 
my_cors <- c(cor(optimal_pred, c(y-2)),
             cor(optimal_pred_1se, c(y-2)), 
            cor(null_pred, c(y-2))
            )

# calculate mse between null and optimal model 
my_mse <- c(mean((fullModOut$modelPredictions$pred_opt -  c(y-2))^2) ,
            mean((fullModOut$modelPredictions$pred_opt_se -  c(y-2))^2) ,
            mean((fullModOut$modelPredictions$pred_null - c(y-2))^2)#,
            #mean((obs_pred$pred_nopenalty - obs_pred$obs)^2)
            )

ggplot() + 
  geom_point(aes(X[,2], fullModOut$modelPredictions$obs-2), col = "black", alpha = .1) + 
  geom_point(aes(X[,2], fullModOut$modelPredictions$pred_opt), col = "red", alpha = .1) + ## predictions w/ the CV model
  geom_point(aes(X[,2], fullModOut$modelPredictions$pred_opt_se), col = "green", alpha = .1) + ## predictions w/ the CV model (1se lambda)
  geom_point(aes(X[,2], fullModOut$modelPredictions$pred_null), col = "blue", alpha = .1) + 
  labs(title = "A rough comparison of observed and model-predicted values", 
       subtitle = "black = observed values \n red = predictions from 'best lambda' model \n green = predictions from '1se' lambda model \n blue = predictions from null model") +
  xlab(colnames(X)[2])

  #ylim(c(0,200))

The internal cross-validation process to fit the global LASSO model identified an optimal lambda value (regularization parameter) of r{print(best_lambda)}. The lambda value such that the cross validation error is within 1 standard error of the minimum (“1se lambda”) was `r{print(fit$lambda.1se)}`` . The following coefficients were kept in each model:

# the coefficient matrix from the 'best model' -- find and print those coefficients that aren't 0 in a table
coef_glm_bestLambda <- coef(fit_glm_bestLambda) %>% 
  data.frame() 
coef_glm_bestLambda$coefficientName <- rownames(coef_glm_bestLambda)
names(coef_glm_bestLambda)[1] <- "coefficientValue_bestLambda"
# coefficient matrix from the '1se' model 
coef_glm_1se <- coef(fit_glm_se) %>% 
  data.frame() 
coef_glm_1se$coefficientName <- rownames(coef_glm_1se)
names(coef_glm_1se)[1] <- "coefficientValue_1seLambda"
# add together
coefs <- full_join(coef_glm_bestLambda, coef_glm_1se) %>% 
  select(coefficientName, coefficientValue_bestLambda, coefficientValue_1seLambda)

globModTerms <- coefs[!is.na(coefs$coefficientValue_bestLambda), "coefficientName"]

## also, get the number of unique variables in each model 
var_prop_pred <- paste0(response, "_pred")
response_vars <- c(response, var_prop_pred)
# for best lambda model
prednames_fig <- paste(str_split(globModTerms, ":", simplify = TRUE)) 
prednames_fig <- str_replace(prednames_fig, "I\\(", "")
prednames_fig <- str_replace(prednames_fig, "\\^2\\)", "")
prednames_fig <- unique(prednames_fig[prednames_fig>0])
prednames_fig <- prednames_fig
prednames_fig_num <- length(prednames_fig)
# for 1SE lambda model
globModTerms_1se <- coefs[!is.na(coefs$coefficientValue_1seLambda), "coefficientName"]
if (length(globModTerms_1se) == 1) {
prednames_fig_1se <- paste(str_split(globModTerms_1se, ":", simplify = TRUE)) 
prednames_fig_1se <- str_replace(prednames_fig_1se, "I\\(", "")
prednames_fig_1se <- str_replace(prednames_fig_1se, "\\^2\\)", "")
prednames_fig_1se <- unique(prednames_fig_1se[prednames_fig_1se>0])
prednames_fig_1se_num <- c(0)
} else {
prednames_fig_1se <- paste(str_split(globModTerms_1se, ":", simplify = TRUE)) 
prednames_fig_1se <- str_replace(prednames_fig_1se, "I\\(", "")
prednames_fig_1se <- str_replace(prednames_fig_1se, "\\^2\\)", "")
prednames_fig_1se <- unique(prednames_fig_1se[prednames_fig_1se>0])
prednames_fig_1se_num <- length(prednames_fig_1se)
}

# make a table
kable(coefs, col.names = c("Coefficient Name", "Value from best lambda model", "Value from 1se lambda model")
      ) %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

Coefficient Name	Value from best lambda model	Value from 1se lambda model
(Intercept)	2.6324176	2.5604940
tmean_s	-0.0987874	NA
prcp_dry_s	-0.1598845	NA
prcpTempCorr_s	-0.2478392	-0.2919912
coarse_s	0.0199590	NA
VPD_mean_anom_s	0.0424590	NA
I(prcp_s^2)	0.0089787	0.0681956
I(prcp_dry_s^2)	-0.0256517	NA
I(precp_dry_anom_s^2)	-0.0016531	NA
I(annWetDegDays_anom_s^2)	-0.0006767	NA
I(clay_s^2)	-0.0167804	-0.0186244
I(AWHC_s^2)	-0.0222037	NA
annWatDef_s:isothermality_anom_s	-0.0096353	NA
prcpTempCorr_s:annWatDef_s	0.0579585	NA
tmean_s:annWatDef_s	0.0442841	NA
annWatDef_s:tmin_anom_s	0.0073024	NA
prcpTempCorr_anom_s:annWetDegDays_anom_s	0.0287549	NA
isothermality_anom_s:isothermality_s	-0.0430421	NA
prcp_dry_s:isothermality_anom_s	0.0164982	NA
prcpTempCorr_s:prcp_s	-0.0737355	NA
tmean_s:VPD_max_95_anom_s	-0.0445939	NA
sand_s:carbon_s	-0.0597064	NA

# calculate RMSE of both models 
RMSE_best <- yardstick::rmse(fullModOut$modelPredictions[,c("obs", "pred_opt")]-2, truth = "obs", estimate = "pred_opt")$.estimate
RMSE_1se <- yardstick::rmse(fullModOut$modelPredictions[,c("obs", "pred_opt_se")]-2, truth = "obs", estimate = "pred_opt_se")$.estimate
bias_best <-  mean(fullModOut$modelPredictions$obs - fullModOut$modelPredictions$pred_opt)
bias_1se <- mean(fullModOut$modelPredictions$obs - fullModOut$modelPredictions$pred_opt_se)

uniqueCoeffs <- data.frame("Best lambda model" = c(RMSE_best, bias_best,
  as.integer(length(globModTerms)-1), as.integer(prednames_fig_num), 
                                                   as.integer(sum(prednames_fig %in% c(prednames_clim))),
                                                   as.integer(sum(prednames_fig %in% c(prednames_weath))),
                                                   as.integer(sum(prednames_fig %in% c(prednames_soils)))
                                                   ), 
                           "1se lambda model" = c(RMSE_1se, bias_1se,
                             length(globModTerms_1se)-1, prednames_fig_1se_num,
                                                   sum(prednames_fig_1se %in% c(prednames_clim)),
                                                   sum(prednames_fig_1se %in% c(prednames_weath)),
                                                   sum(prednames_fig_1se %in% c(prednames_soils))))
row.names(uniqueCoeffs) <- c("RMSE", "bias - mean(obs-pred.)", "Total number of coefficients", "Number of unique coefficients",
                             "Number of unique climate coefficients", 
                             "Number of unique weather coefficients",  
                             "Number of unique soils coefficients"
                             )

kable(uniqueCoeffs, 
      col.names = c("Best lambda model", "1se lambda model"), row.names = TRUE) %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

	Best lambda model	1se lambda model
RMSE	16.064214	16.757923
bias - mean(obs-pred.)	2.030205	1.916181
Total number of coefficients	21.000000	3.000000
Number of unique coefficients	18.000000	3.000000
Number of unique climate coefficients	6.000000	2.000000
Number of unique weather coefficients	7.000000	0.000000
Number of unique soils coefficients	5.000000	1.000000

Visualizations of Model Predictions and Residuals – using best lambda model

observed vs. predicted values

Predicting on the data

  # create prediction for each each model
# (i.e. for each fire proporation variable)
predict_by_response <- function(mod, df) {
  df_out <- df
  response_name <- paste0(response, "_pred")
  df_out <- df_out %>% cbind(predict(mod, newx= df_out, #s="lambda.min", 
                                     type = "response"))
   colnames(df_out)[ncol(df_out)] <- response_name
  return(df_out)
}

pred_glm1 <- predict_by_response(fit_glm_bestLambda, X[,2:ncol(X)])

# add back in true y values
pred_glm1 <- pred_glm1 %>% 
  cbind( data.frame("y" = y))
# rename the true response column to not be 'y_test' 
colnames(pred_glm1)[which(colnames(pred_glm1) == "y")] <- paste(response)

# add back in lat/long data 
pred_glm1 <- pred_glm1 %>% 
  cbind(modDat_1_s[,c("Long", "Lat", "Year")])

pred_glm1$resid <- pred_glm1[,response] - pred_glm1[,paste0(response, "_pred")]
pred_glm1$extremeResid <- NA
pred_glm1[pred_glm1$resid > 70 | pred_glm1$resid < -70,"extremeResid"] <- 1

# plot(x = pred_glm1[,response],
#      y = pred_glm1[,paste0(response, "_pred")],
#      xlab = "observed values", ylab = "predicted values")
# points(x = pred_glm1[!is.na(pred_glm1$extremeResid), response],
#        y = pred_glm1[!is.na(pred_glm1$extremeResid), paste0(response, "_pred")],
#        col = "red")

pred_glm1_1se <- predict_by_response(fit_glm_se, X[,2:ncol(X)])

# add back in true y values
pred_glm1_1se <- pred_glm1_1se %>% 
  cbind( data.frame("y" = y))
# rename the true response column to not be 'y_test' 
colnames(pred_glm1_1se)[which(colnames(pred_glm1_1se) == "y")] <- paste(response)

# add back in lat/long data 
pred_glm1_1se <- pred_glm1_1se %>% 
  cbind(modDat_1_s[,c("Long", "Lat", "Year")])

pred_glm1_1se$resid <- pred_glm1_1se[,response] - pred_glm1_1se[,paste0(response, "_pred")]
pred_glm1_1se$extremeResid <- NA
pred_glm1_1se[pred_glm1_1se$resid > 70 | pred_glm1_1se$resid < -70,"extremeResid"] <- 1

Maps of Observations, Predictions, and Residuals=

Observations across the temporal range of the dataset

pred_glm1 <- pred_glm1 %>% 
  mutate(resid = .[[response]] - .[[paste0(response,"_pred")]]) 

# rasterize
# get reference raster
test_rast <-  rast("../../../Data_raw/dayMet/rawMonthlyData/orders/70e0da02b9d2d6e8faa8c97d211f3546/Daymet_Monthly_V4R1/data/daymet_v4_prcp_monttl_na_1980.tif") %>% 
  terra::aggregate(fact = 8, fun = "mean")

## |---------|---------|---------|---------|=========================================                                          

## add ecoregion boundaries (for our ecoregion level model)
regions <- sf::st_read(dsn = "../../../Data_raw/Level2Ecoregions/", layer = "NA_CEC_Eco_Level2") 

## Reading layer `NA_CEC_Eco_Level2' from data source 
##   `/Users/astears/Documents/Dropbox_static/Work/NAU_USGS_postdoc/PED_vegClimModels/Data_raw/Level2Ecoregions' using driver `ESRI Shapefile'
## Simple feature collection with 2261 features and 8 fields
## Geometry type: POLYGON
## Dimension:     XY
## Bounding box:  xmin: -4334052 ymin: -3313739 xmax: 3324076 ymax: 4267265
## Projected CRS: Sphere_ARC_INFO_Lambert_Azimuthal_Equal_Area

regions <- regions %>% 
  st_transform(crs = st_crs(test_rast)) %>% 
  st_make_valid() #%>% 
  #st_crop(st_bbox(test_rast))
# 
# goodRegions_temp <- st_overlaps(y = cropped_states, x = regions, sparse = FALSE) %>% 
#   rowSums() 
# goodRegions <- regions[goodRegions_temp != 0,]

ecoregionLU <- data.frame("NA_L1NAME" = sort(unique(regions$NA_L1NAME)), 
                        "newRegion" = c(NA, "Forest", "dryShrubGrass", 
                                        "dryShrubGrass", "Forest", "dryShrubGrass",
                                       "dryShrubGrass", "Forest", "Forest", 
                                       "dryShrubGrass", "Forest", "Forest", 
                                       "Forest", "Forest", "dryShrubGrass", 
                                       NA
                                        ))
goodRegions <- regions %>% 
  left_join(ecoregionLU)
mapRegions <- goodRegions %>% 
  filter(!is.na(newRegion)) %>% 
  group_by(newRegion) %>% 
  summarise(geometry = sf::st_union(geometry)) %>% 
  ungroup() %>% 
  st_simplify(dTolerance = 1000)
#mapview(mapRegions)
# rasterize data
plotObs <- pred_glm1 %>% 
         drop_na(paste(response)) %>% 
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) %>% 
  terra::rasterize(y = test_rast, 
                   field = response, 
                   fun = mean) #%>% 
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>% 
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

# get the extent of this particular raster, and crop it accordingly
tempExt <- crds(plotObs, na.rm = TRUE)

plotObs_2 <- plotObs %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )
# make figures
ggplot() +
geom_spatraster(data = plotObs_2) + 
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA ) +
labs(title = paste0("Observations of ", response, " in the ",ecoregion, " ecoregion")) +
  scale_fill_gradient2(low = "brown",
                       mid = "wheat" ,
                       high = "darkgreen" , 
                       midpoint = 0,   na.value = "lightgrey") + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])

Predictions across the temporal range of the dataset

# rasterize data
plotPred <- pred_glm1 %>% 
         drop_na(paste0(response,"_pred")) %>% 
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) %>% 
  terra::rasterize(y = test_rast, 
                   field = paste0(response,"_pred"), 
                   fun = mean) #%>% 
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>% 
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

# get the point location of those predictions that are > 100
highPred_points <- pred_glm1 %>% 
  filter(.[[paste0(response,"_pred")]] > 100 | 
           .[[paste0(response, "_pred")]] < 0) %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 

# get the extent of this particular raster, and crop it accordingly
tempExt <- crds(plotPred, na.rm = TRUE)

plotPred_2 <- plotPred %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )
# make figures
ggplot() +
geom_spatraster(data = plotPred_2) + 
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA )  + 
  geom_sf(data = highPred_points, col = "red") +
labs(title = paste0("Predictions from the 'best lambda' fitted model of ", response, " in the ",ecoregion, " ecoregion"),
     subtitle =  "bestLambda model")  +
  scale_fill_gradient2(low = "wheat",
                       mid = "darkgreen",
                       high = "red" , 
                       midpoint = 100,   na.value = "lightgrey",
                       limits = c(0,100)) + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])

# rasterize data
plotPred <- pred_glm1_1se %>% 
         drop_na(paste0(response,"_pred")) %>% 
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) %>% 
  terra::rasterize(y = test_rast, 
                   field = paste0(response,"_pred"), 
                   fun = mean) #%>% 
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>% 
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

# get the point location of those predictions that are > 100
highPred_points <- pred_glm1_1se %>% 
  filter(.[[paste0(response,"_pred")]] > 100 | 
           .[[paste0(response, "_pred")]] < 0) %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 

# get the extent of this particular raster, and crop it accordingly
tempExt <- crds(plotPred, na.rm = TRUE)

plotPred_2 <- plotPred %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )
# make figures
ggplot() +
geom_spatraster(data = plotPred_2) + 
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA )  + geom_sf(data = highPred_points, col = "red") +
labs(title = paste0("Predictions from the '1SE lambda' fitted model of ", response, " in the ",ecoregion, " ecoregion"),
     subtitle =  "1 SE Lambda model")  +
  scale_fill_gradient2(low = "wheat",
                       mid = "darkgreen",
                       high = "red" , 
                       midpoint = 100,   na.value = "lightgrey",
                       limits = c(0,100)) + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])

Residuals across the entire temporal extent of the dataset

# rasterize data
plotResid_rast <- pred_glm1 %>% 
         drop_na(resid) %>% 
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) %>% 
  terra::rasterize(y = test_rast, 
                   field = "resid", 
                   fun = mean) #%>% 
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>% 
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

# get the extent of this particular raster, and crop it accordingly
tempExt <- crds(plotResid_rast, na.rm = TRUE)

plotResid_rast_2 <- plotResid_rast %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )

# identify locations where residuals are >100 or < -100
badResids_high <- pred_glm1 %>% 
  filter(resid > 100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 
badResids_low <- pred_glm1 %>% 
  filter(resid < -100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 
# make figures
map <- ggplot() +
geom_spatraster(data =plotResid_rast_2) + 
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA )  + 
  geom_sf(data = badResids_high, col = "blue") +
  geom_sf(data = badResids_low, col = "red") +
labs(title = paste0("Resids. (obs. - pred.) from Grass/shrub ecoregion-wide model of ", response),
     subtitle = "bestLambda model \n red points indicate locations that have residuals below -100 \n blue points indicate locatiosn that have residuals above 100") +
  scale_fill_gradient2(low = "red",
                       mid = "white" ,
                       high = "blue" , 
                       midpoint = 0,   na.value = "lightgrey",
                       limits = c(-100,100)
                       ) + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])
hist <- ggplot(pred_glm1) + 
  geom_histogram(aes(resid), fill = "lightgrey", col = "darkgrey") + 
  geom_text(aes(x = min(resid)*.9, y = 1500, label = paste0("min = ", round(min(resid),2)))) +
  geom_text(aes(x = max(resid)*.9, y = 1500, label = paste0("max = ", round(max(resid),2))))

library(ggpubr)
ggarrange(map, hist, heights = c(3,1), ncol = 1)

# rasterize data
plotResid_rast <- pred_glm1_1se %>% 
         drop_na(resid) %>% 
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) %>% 
  terra::rasterize(y = test_rast, 
                   field = "resid", 
                   fun = mean) #%>% 
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>% 
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

# get the extent of this particular raster, and crop it accordingly
tempExt <- crds(plotResid_rast, na.rm = TRUE)

plotResid_rast_2 <- plotResid_rast %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )

# identify locations where residuals are >100 or < -100
badResids_high <- pred_glm1_1se %>% 
  filter(resid > 100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 
badResids_low <- pred_glm1_1se %>% 
  filter(resid < -100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 
# make figures
map <- ggplot() +
geom_spatraster(data =plotResid_rast_2) + 
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA )  + 
  geom_sf(data = badResids_high, col = "blue") +
  geom_sf(data = badResids_low, col = "red") +
labs(title = paste0("Resids. (obs. - pred.) from Grass/shrub ecoregion-wide model of ", response),
     subtitle = "1 SE Lambda model \n red points indicate locations that have residuals below -100 \n blue points indicate locatiosn that have residuals above 100") +
  scale_fill_gradient2(low = "red",
                       mid = "white" ,
                       high = "blue" , 
                       midpoint = 0,   na.value = "lightgrey",
                       limits = c(-100,100)
                       ) + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])
hist <- ggplot(pred_glm1_1se) + 
  geom_histogram(aes(resid), fill = "lightgrey", col = "darkgrey") + 
  geom_text(aes(x = min(resid)*.9, y = 1500, label = paste0("min = ", round(min(resid),2)))) +
  geom_text(aes(x = max(resid)*.9, y = 1500, label = paste0("max = ", round(max(resid),2))))

ggarrange(map, hist, heights = c(3,1), ncol = 1)

### Are there biases of the model predictions across year/lat/long?

# plot residuals against Year
yearResidMod_bestLambda <- ggplot(pred_glm1) + 
  geom_point(aes(x = jitter(Year), y = resid), alpha = .1) + 
  geom_smooth(aes(x = Year, y = resid)) + 
  xlab("Year") + 
  ylab("Residual from best lambda model") +
  ggtitle("from best lamba model")
yearResidMod_1seLambda <- ggplot(pred_glm1_1se) + 
  geom_point(aes(x = jitter(Year), y = resid), alpha = .1) + 
  geom_smooth(aes(x = Year, y = resid)) + 
  xlab("Year") + 
  ylab("Residual from 1 SE lambda model")+
  ggtitle("from 1 SE lamba model")

# plot residuals against Lat
latResidMod_bestLambda <- ggplot(pred_glm1) + 
  geom_point(aes(x = Lat, y = resid), alpha = .1) + 
  geom_smooth(aes(x = Lat, y = resid)) + 
  xlab("Latitude") + 
  ylab("Residual from best lambda model") +
  ggtitle("from best lamba model")
latResidMod_1seLambda <- ggplot(pred_glm1_1se) + 
  geom_point(aes(x = Lat, y = resid), alpha = .1) + 
  geom_smooth(aes(x = Lat, y = resid)) + 
  xlab("Latitude") + 
  ylab("Residual from 1 SE lambda model") +
  ggtitle("from 1 SE lamba model")

# plot residuals against Long
longResidMod_bestLambda <- ggplot(pred_glm1) + 
  geom_point(aes(x = Long, y = resid), alpha = .1) + 
  geom_smooth(aes(x = Long, y = resid)) + 
  xlab("Longitude") + 
  ylab("Residual from best lambda model") +
  ggtitle("from best lamba model")
longResidMod_1seLambda <- ggplot(pred_glm1_1se) + 
  geom_point(aes(x = Long, y = resid), alpha = .1) + 
  geom_smooth(aes(x = Long, y = resid)) + 
  xlab("Longitude") + 
  ylab("Residual from 1 SE lambda model") +
  ggtitle("from 1 SE lamba model")

library(patchwork)
(yearResidMod_bestLambda + yearResidMod_1seLambda) / 
(  latResidMod_bestLambda + latResidMod_1seLambda) /
(  longResidMod_bestLambda + longResidMod_1seLambda)

Quantile plots

Binning predictor variables into “Deciles” (actually percentiles) and looking at the mean predicted probability for each percentile. The use of the word Decentiles is just a legacy thing (they started out being actual Percentiles)

# get deciles for best lambda model 
if (length(prednames_fig) == 0) {
  print("The best lambda model only contains one predictor (an intercept), so decile plots aren't possible to generate")
} else {
  pred_glm1_deciles <- predvars2deciles(pred_glm1,
                                      response_vars = response_vars,
                                        pred_vars = prednames_fig, 
                                       cut_points = seq(0, 1, 0.005))
}

## Processed 3304 groups out of 3577. 92% done. Time elapsed: 3s. ETA: 0s.Processed 3577 groups out of 3577. 100% done. Time elapsed: 3s. ETA: 0s.

# get deciles for 1 SE lambda model 
if (length(prednames_fig_1se) == 0) {
  print("The 1SE lambda model only contains one predictor (an intercept), so decile plots aren't possible to generate")
} else {
  pred_glm1_deciles_1se <- predvars2deciles(pred_glm1_1se,
                                      response_vars = response_vars,
                                        pred_vars = prednames_fig_1se, 
                                       cut_points = seq(0, 1, 0.005))
}

Here is a quick version of LOESS curves fit to raw data (to double-check the quantile plot calculations)

if (length(prednames_fig) == 0) {
  print("The model only contains one predictor (an intercept), so decile plots aren't possible to generate")
} else {
  pred_glm1 %>%
  select(all_of(c(prednames_fig, response_vars))) %>%
  pivot_longer(cols = prednames_fig)  %>%
  ggplot() +
  facet_wrap(~name, scales = "free") +
  geom_point(aes(x = value, y =  .data[[paste(response)]]), col = "darkblue", alpha = .1)  + # observed values
  geom_point(aes(x = value, y = .data[[response_vars[2]]]), col = "lightblue", alpha = .1) + # model-predicted values
  geom_smooth(aes(x = value, y =  .data[[paste(response)]]), col = "black", se = FALSE) +
  geom_smooth(aes(x = value, y = .data[[response_vars[2]]]), col = "brown", se = FALSE)

}

Below are the actual quantile plots (note that the predictor variables are scaled)

if (length(prednames_fig) == 0) {
  print("The model only contains one predictor (an intercept), so decile plots aren't possible to generate")
} else {

# publication quality version
g3 <- decile_dotplot_pq(pred_glm1_deciles, response = response, IQR = TRUE) + ggtitle("Decile Plot")

g4 <- add_dotplot_inset(g3, df = pred_glm1_deciles, dfRaw = pred_glm1, add_smooth = TRUE, deciles = FALSE)

  
if(save_figs) {
  png(paste0("figures/quantile_plots/quantile_plot_", response,  "_",ecoregion,".png"), 
     units = "in", res = 600, width = 5.5, height = 3.5 )
    print(g4)
  dev.off()
}

g4
}

if (length(prednames_fig_1se) == 0) {
  print("The 1 se lambda model only contains one predictor (an intercept), so decile plots aren't possible to generate")

  } else {

# publication quality version
g3 <- decile_dotplot_pq(pred_glm1_deciles_1se, response = response, IQR = TRUE) + ggtitle("Decile Plot")

g4 <- add_dotplot_inset(g3, df = pred_glm1_deciles_1se, dfRaw = pred_glm1_1se, add_smooth = TRUE, deciles = FALSE)

  
if(save_figs) {
  png(paste0("figures/quantile_plots/quantile_plot_", response,  "_",ecoregion,".png"), 
     units = "in", res = 600, width = 5.5, height = 3.5 )
    print(g4)
  dev.off()
}

g4
}

Deciles Filtered

20th and 80th percentiles for each climate variable

df <- pred_glm1[, prednames_fig] #%>% 
  #mutate(MAT = MAT - 273.15) # k to c
quantiles <- map(df, quantile, probs = c(0.2, 0.8), na.rm = TRUE)

Filtered ‘Decile’ plots of data. These plots show each vegetation variable, but only based on data that falls into the upper and lower two deciles of each predictor variable.

if (length(prednames_fig) == 0) {
  print("The model only contains one predictor (an intercept), so decile plots aren't possible to generate")
} else {
pred_glm1_deciles_filt <- predvars2deciles( pred_glm1, 
                         response_vars = response_vars,
                         pred_vars = prednames_fig,
                         filter_var = TRUE,
                         filter_vars = prednames_fig,
                         cut_points = seq(0, 1, 0.005)) 

decile_dotplot_filtered_pq(pred_glm1_deciles_filt, xvars = prednames_fig)
#decile_dotplot_filtered_pq(pred_glm1_deciles_filt)

}

## Processed 16037 groups out of 128526. 12% done. Time elapsed: 3s. ETA: 21s.Processed 20495 groups out of 128526. 16% done. Time elapsed: 4s. ETA: 23s.Processed 25826 groups out of 128526. 20% done. Time elapsed: 5s. ETA: 21s.Processed 31263 groups out of 128526. 24% done. Time elapsed: 6s. ETA: 19s.Processed 36764 groups out of 128526. 29% done. Time elapsed: 7s. ETA: 18s.Processed 42259 groups out of 128526. 33% done. Time elapsed: 8s. ETA: 17s.Processed 47809 groups out of 128526. 37% done. Time elapsed: 9s. ETA: 15s.Processed 53183 groups out of 128526. 41% done. Time elapsed: 10s. ETA: 14s.Processed 58589 groups out of 128526. 46% done. Time elapsed: 11s. ETA: 13s.Processed 64016 groups out of 128526. 50% done. Time elapsed: 12s. ETA: 12s.Processed 69437 groups out of 128526. 54% done. Time elapsed: 13s. ETA: 11s.Processed 74931 groups out of 128526. 58% done. Time elapsed: 14s. ETA: 10s.Processed 80445 groups out of 128526. 63% done. Time elapsed: 15s. ETA: 9s.Processed 85940 groups out of 128526. 67% done. Time elapsed: 16s. ETA: 8s.Processed 91463 groups out of 128526. 71% done. Time elapsed: 17s. ETA: 7s.Processed 96988 groups out of 128526. 75% done. Time elapsed: 18s. ETA: 5s.Processed 102402 groups out of 128526. 80% done. Time elapsed: 19s. ETA: 4s.Processed 107529 groups out of 128526. 84% done. Time elapsed: 20s. ETA: 3s.Processed 112880 groups out of 128526. 88% done. Time elapsed: 21s. ETA: 2s.Processed 118341 groups out of 128526. 92% done. Time elapsed: 22s. ETA: 1s.Processed 123820 groups out of 128526. 96% done. Time elapsed: 23s. ETA: 0s.Processed 128526 groups out of 128526. 100% done. Time elapsed: 24s. ETA: 0s.

Filtered quantile figure with middle 2 deciles also shown

if (length(prednames_fig) == 0) {
  print("The model only contains one predictor (an intercept), so decile plots aren't possible to generate")
} else {
pred_glm1_deciles_filt_mid <- predvars2deciles(pred_glm1, 
                         response_vars = response_vars,
                         pred_vars = prednames_fig,
                         filter_vars = prednames_fig,
                         filter_var = TRUE,
                         add_mid = TRUE,
                         cut_points = seq(0, 1, 0.005))

g <- decile_dotplot_filtered_pq(df = pred_glm1_deciles_filt_mid, xvars = prednames_fig)
g

if(save_figs) {x
jpeg(paste0("figures/quantile_plots/quantile_plot_filtered_mid_v1", , ".jpeg"),
     units = "in", res = 600, width = 5.5, height = 6 )
  g 
dev.off()
}
}

## Processed 15904 groups out of 192953. 8% done. Time elapsed: 3s. ETA: 33s.Processed 21346 groups out of 192953. 11% done. Time elapsed: 4s. ETA: 32s.Processed 26672 groups out of 192953. 14% done. Time elapsed: 5s. ETA: 31s.Processed 32102 groups out of 192953. 17% done. Time elapsed: 6s. ETA: 30s.Processed 37473 groups out of 192953. 19% done. Time elapsed: 7s. ETA: 29s.Processed 42783 groups out of 192953. 22% done. Time elapsed: 8s. ETA: 28s.Processed 48139 groups out of 192953. 25% done. Time elapsed: 9s. ETA: 27s.Processed 53540 groups out of 192953. 28% done. Time elapsed: 10s. ETA: 26s.Processed 58792 groups out of 192953. 30% done. Time elapsed: 11s. ETA: 25s.Processed 64152 groups out of 192953. 33% done. Time elapsed: 12s. ETA: 24s.Processed 69463 groups out of 192953. 36% done. Time elapsed: 13s. ETA: 23s.Processed 74858 groups out of 192953. 39% done. Time elapsed: 14s. ETA: 22s.Processed 79518 groups out of 192953. 41% done. Time elapsed: 15s. ETA: 22s.Processed 85045 groups out of 192953. 44% done. Time elapsed: 16s. ETA: 21s.Processed 90670 groups out of 192953. 47% done. Time elapsed: 17s. ETA: 20s.Processed 96243 groups out of 192953. 50% done. Time elapsed: 18s. ETA: 18s.Processed 101828 groups out of 192953. 53% done. Time elapsed: 19s. ETA: 17s.Processed 107396 groups out of 192953. 56% done. Time elapsed: 20s. ETA: 16s.Processed 112901 groups out of 192953. 59% done. Time elapsed: 21s. ETA: 15s.Processed 118472 groups out of 192953. 61% done. Time elapsed: 22s. ETA: 14s.Processed 124051 groups out of 192953. 64% done. Time elapsed: 23s. ETA: 13s.Processed 129662 groups out of 192953. 67% done. Time elapsed: 24s. ETA: 12s.Processed 135162 groups out of 192953. 70% done. Time elapsed: 25s. ETA: 11s.Processed 140756 groups out of 192953. 73% done. Time elapsed: 26s. ETA: 9s.Processed 146202 groups out of 192953. 76% done. Time elapsed: 27s. ETA: 8s.Processed 151430 groups out of 192953. 78% done. Time elapsed: 28s. ETA: 7s.Processed 156703 groups out of 192953. 81% done. Time elapsed: 29s. ETA: 6s.Processed 161942 groups out of 192953. 84% done. Time elapsed: 30s. ETA: 5s.Processed 167364 groups out of 192953. 87% done. Time elapsed: 31s. ETA: 4s.Processed 172943 groups out of 192953. 90% done. Time elapsed: 32s. ETA: 3s.Processed 178532 groups out of 192953. 93% done. Time elapsed: 33s. ETA: 2s.Processed 184084 groups out of 192953. 95% done. Time elapsed: 34s. ETA: 1s.Processed 189435 groups out of 192953. 98% done. Time elapsed: 35s. ETA: 0s.Processed 192953 groups out of 192953. 100% done. Time elapsed: 36s. ETA: 0s.

Cross-validation

Using best lambda model

Use terms from global model to re-fit and predict on different held out regions

Figures show residuals for each of the models fit to held-out ecoregions

These models were fit to six ecoregions, and then predict on the indicated heldout ecoregion

if (length(prednames_fig) == 0) {
  print("The model only contains one predictor (an intercept), so cross validation isn't practical")
} else {
  
## code from Tredennick et al. 2020
# try each separate level II ecoregion as a test set
# make a list to hold output data
outList <- vector(mode = "list", length = length(sort(unique(modDat_1$NA_L2NAME))))
# obs_pred <- data.frame(ecoregion = character(),obs = numeric(),
#                        pred_opt = numeric(), pred_null = numeric()#,
#                        #pred_nopenalty = numeric()
#                        )

## get the model specification from the global model
mat <- as.matrix(coef(fit_glm_bestLambda, s = "lambda.min"))
mat2 <- mat[mat[,1] != 0,]

f_cv <- as.formula(paste0(response, " ~ ", paste0(names(mat2)[2:length(names(mat2))], collapse = " + ")))

X_cv <- model.matrix(object = f_cv, data = modDat_1_s)
# get response variable
y_cv <- as.matrix(modDat_1_s[,response])

  
# now, loop through so with each iteration, a different ecoregion is held out
 for(i_eco in sort(unique(modDat_1_s$NA_L2NAME))){

  # split into training and test sets
  test_eco <- i_eco
  print(test_eco)
  # identify the rowID of observations to be in the training and test datasets
  train <- which(modDat_1_s$NA_L2NAME!=test_eco) # data for all ecoregions that aren't 'i_eco'
  test <- which(modDat_1_s$NA_L2NAME==test_eco) # data for the ecoregion that is 'i_eco'

  trainDat_all <- modDat_1_s %>% 
    slice(train) %>% 
    select(-newRegion)
  testDat_all <- modDat_1_s %>% 
    slice(test) %>% 
    select(-newRegion)

  # get the model matrices for input and response variables for cross validation model specification
  X_train <- as.matrix(X_cv[train,])
  X_test <- as.matrix(X_cv[test,])

  y_train <- modDat_1_s[train,response]
  y_test <- modDat_1_s[test,response]
  
  # get the model matrices for input and response variables for original model specification
  X_train_glob <- as.matrix(X[train,])
  X_test_glob <- as.matrix(X[test,])

  y_train_glob <- modDat_1_s[train,response]
  y_test_glob <- modDat_1_s[test,response]

  train_eco <- modDat_1_s$NA_L2NAME[train]

  ## just try a regular glm w/ the components from the global model
  fit_i <- glm(data = trainDat_all, formula = f_cv, 
    ,
               family =  stats::Gamma(link = "log")
    )
    
  # lasso model predictions with the optimal lambda
  optimal_pred <- predict(fit_i, newdata= testDat_all, type = "response"
                          )
  # null model and predictions
  # the "null" model in this case is the global model
  # predict on the test data for this iteration w/ the global model 
  null_pred <- predict.glm(fit_glm_bestLambda, newdata = testDat_all, type = "response")

  
  # save data
  tmp <- data.frame(ecoRegion_holdout = rep(test_eco,length(y_test)),obs=y_test,
                    pred_opt=optimal_pred, pred_null=null_pred#,
                    #pred_nopenalty=nopen_pred
                    ) %>%
    cbind(testDat_all)
  
  # calculate RMSE, bias, etc. of 
  # RMSE of CV model 
  RMSE_optimal <- yardstick::rmse(data = data.frame(optimal_pred, y_test), truth = "y_test", estimate = "optimal_pred")[1,]$.estimate
  # RMSE of global model
  RMSE_null <- yardstick::rmse(data = data.frame(null_pred, y_test), truth = "y_test", estimate = "null_pred")[1,]$.estimate
  # bias of CV model
  bias_optimal <- mean(y_test - optimal_pred)
  # bias of global model
  bias_null <-  mean( y_test - null_pred )
  
  # put output into a list
  tmpList <- list("testRegion" = i_eco,
    "modelObject" = fit_i,
       "modelPredictions" = tmp, 
    "performanceMetrics" = data.frame("RMSE_cvModel" = RMSE_optimal, 
                                      "RMSE_globalModel" = RMSE_null, 
                                      "bias_cvModel" = bias_optimal, 
                                      "bias_globalModel" = bias_null))

  # save model outputs
  outList[[which(sort(unique(modDat_1_s$NA_L2NAME)) == i_eco)]] <- tmpList
 }
}

## [1] "ATLANTIC HIGHLANDS"
## [1] "CENTRAL USA PLAINS"
## [1] "EVERGLADES"
## [1] "MARINE WEST COAST FOREST"
## [1] "MISSISSIPPI ALLUVIAL AND SOUTHEAST USA COASTAL PLAINS"
## [1] "MIXED WOOD PLAINS"
## [1] "MIXED WOOD SHIELD"
## [1] "OZARK/OUACHITA-APPALACHIAN FORESTS"
## [1] "SOUTHEASTERN USA PLAINS"
## [1] "UPPER GILA MOUNTAINS"
## [1] "WESTERN CORDILLERA"

Below are the RMSE and bias values for predictions made for each holdout level II ecoregion, compared to predictions from the global model for that same ecoregion

# table of model performance
map(outList, .f = function(x) {
  cbind(data.frame("holdout region" = x$testRegion),  x$performanceMetrics)
}
) %>% 
  purrr::list_rbind() %>% 
  kable(col.names = c("Held-out ecoregion", "RMSE of CV model", "RMSE of global model", 
                      "bias of CV model - mean(obs-pred.)", "bias of global model- mean(obs-pred.)"), 
        caption = "Performance of Cross Validation using 'best lambda' model specification") %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 

Performance of Cross Validation using ‘best lambda’ model specification

Held-out ecoregion	RMSE of CV model	RMSE of global model	bias of CV model - mean(obs-pred.)	bias of global model- mean(obs-pred.)
ATLANTIC HIGHLANDS	11.425531	11.389826	-0.9366204	-0.7758429
CENTRAL USA PLAINS	14.301578	14.296019	1.2531566	1.2199825
EVERGLADES	6.682387	6.521044	-3.5478916	-3.2669659
MARINE WEST COAST FOREST	28.297481	27.619419	2.2295604	2.4965971
MISSISSIPPI ALLUVIAL AND SOUTHEAST USA COASTAL PLAINS	13.964923	13.864107	2.0221372	1.0981103
MIXED WOOD PLAINS	12.292830	12.257960	-1.0809792	-0.8498037
MIXED WOOD SHIELD	14.649890	14.636564	0.2033273	-0.0046839
OZARK/OUACHITA-APPALACHIAN FORESTS	13.529430	13.497629	-0.1788398	-0.1736255
SOUTHEASTERN USA PLAINS	11.743247	11.415789	-2.7296333	-1.1381447
UPPER GILA MOUNTAINS	12.010682	11.521479	-2.6740377	-1.5662377
WESTERN CORDILLERA	18.237259	17.556579	2.8905285	0.4103751

# visualize model predictions
for (i in 1:length(unique(modDat_1_s$NA_L2NAME))) {
  holdoutRegion <- outList[[i]]$testRegion
  predictionData <- outList[[i]]$modelPredictions
  modTerms <- as.matrix(coef(outList[[i]]$modelObject)) %>%
    as.data.frame() %>%
    filter(V1!=0) %>%
    rownames()

  # calculate residuals
  predictionData <- predictionData %>%
  mutate(resid = .[["obs"]] - .[["pred_opt"]] ,
         resid_globMod = .[["obs"]]  - .[["pred_null"]])


# rasterize
# use 'test_rast' from earlier

  # rasterize data
plotObs <- predictionData %>%
         drop_na(paste(response)) %>%
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>%
  terra::set.crs(crs(test_rast)) %>%
  terra::rasterize(y = test_rast,
                   field = "resid",
                   fun = mean) #%>%
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>%
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

tempExt <- crds(plotObs, na.rm = TRUE)

plotObs_2 <- plotObs %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )

# identify locations where residuals are >100 or < -100
badResids_high <- predictionData %>% 
  filter(resid > 100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 
badResids_low <- predictionData %>% 
  filter(resid < -100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 


# make figures
# make histogram
hist_i <- ggplot(predictionData) +
  geom_histogram(aes(resid_globMod), col = "darkgrey", fill = "lightgrey") +
  xlab(c("Residuals (obs. - pred.)"))
# make map
map_i <-  ggplot() +
geom_spatraster(data = plotObs_2) +
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA ) +
  geom_sf(data = badResids_high, col = "blue") +
  geom_sf(data = badResids_low, col = "red") +
labs(title = paste0("Residuals (obs. - pred.) for predictions of \n", holdoutRegion, " \n from a model fit to other ecoregions"),
     subtitle = paste0(response, " ~ ", paste0( modTerms, collapse = " + "))) +
  scale_fill_gradient2(low = "red",
                       mid = "white" ,
                       high = "blue" ,
                       midpoint = 0,   na.value = "lightgrey",
                       limits = c(-100, 100))  + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])

 assign(paste0("residPlot_",holdoutRegion),
   value = ggarrange(map_i, hist_i, heights = c(3,1), ncol = 1)
)

}

  lapply(unique(modDat_1_s$NA_L2NAME), FUN = function(x) {
    get(paste0("residPlot_", x))
  })

## [[1]]

## 
## [[2]]

## 
## [[3]]

## 
## [[4]]

## 
## [[5]]

## 
## [[6]]

## 
## [[7]]

## 
## [[8]]

## 
## [[9]]

## 
## [[10]]

## 
## [[11]]

Using 1se lambda model

Use terms from global model to re-fit and predict on different held out regions

Figures show residuals for each of the models fit to held-out ecoregions

These models were fit to six ecoregions, and then predict on the indicated heldout ecoregion

if (length(prednames_fig_1se) == 0) {
  print("The model only contains one predictor (an intercept), so cross validation isn't practical")
} else {

## code from Tredennick et al. 2020
# try each separate level II ecoregion as a test set
# make a list to hold output data
outList <- vector(mode = "list", length = length(sort(unique(modDat_1$NA_L2NAME))))
# obs_pred <- data.frame(ecoregion = character(),obs = numeric(),
#                        pred_opt = numeric(), pred_null = numeric()#,
#                        #pred_nopenalty = numeric()
#                        )

## get the model specification from the global model
mat <- as.matrix(coef(fit_glm_se, s = "lambda.min"))
mat2 <- mat[mat[,1] != 0,]

f_cv <- as.formula(paste0(response, " ~ ", paste0(names(mat2)[2:length(names(mat2))], collapse = " + ")))

X_cv <- model.matrix(object = f_cv, data = modDat_1_s)
# get response variable
y_cv <- as.matrix(modDat_1_s[,response])

  
# now, loop through so with each iteration, a different ecoregion is held out
 for(i_eco in sort(unique(modDat_1_s$NA_L2NAME))){

  # split into training and test sets
  test_eco <- i_eco
  print(test_eco)
  # identify the rowID of observations to be in the training and test datasets
  train <- which(modDat_1_s$NA_L2NAME!=test_eco) # data for all ecoregions that aren't 'i_eco'
  test <- which(modDat_1_s$NA_L2NAME==test_eco) # data for the ecoregion that is 'i_eco'

  trainDat_all <- modDat_1_s %>% 
    slice(train) %>% 
    select(-newRegion)
  testDat_all <- modDat_1_s %>% 
    slice(test) %>% 
    select(-newRegion)

  # get the model matrices for input and response variables for cross validation model specification
  X_train <- as.matrix(X_cv[train,])
  X_test <- as.matrix(X_cv[test,])

  y_train <- modDat_1_s[train,response]
  y_test <- modDat_1_s[test,response]
  
  # get the model matrices for input and response variables for original model specification
  X_train_glob <- as.matrix(X[train,])
  X_test_glob <- as.matrix(X[test,])

  y_train_glob <- modDat_1_s[train,response]
  y_test_glob <- modDat_1_s[test,response]

  train_eco <- modDat_1_s$NA_L2NAME[train]

  ## just try a regular glm w/ the components from the global model
  fit_i <- glm(data = trainDat_all, formula = f_cv, 
    ,
               family =  stats::Gamma(link = "log")
    )

    coef(fit_i)
    
  # lasso model predictions with the optimal lambda
  optimal_pred <- predict(fit_i, newdata= testDat_all, type = "response"
                          )
  # null model and predictions
  # the "null" model in this case is the global model
  # predict on the test data for this iteration w/ the global model 
  null_pred <- predict.glm(fit_glm_se, newdata = testDat_all, type = "response")

  # save data
  tmp <- data.frame(ecoRegion_holdout = rep(test_eco,length(y_test)),obs=y_test,
                    pred_opt=optimal_pred, pred_null=null_pred#,
                    #pred_nopenalty=nopen_pred
                    ) %>%
    cbind(testDat_all)
    
  # calculate RMSE, bias, etc. of 
  # RMSE of CV model 
  RMSE_optimal <- yardstick::rmse(data = data.frame(optimal_pred, y_test), truth = "y_test", estimate = "optimal_pred")[1,]$.estimate
  # RMSE of global model
  RMSE_null <- yardstick::rmse(data = data.frame(null_pred, y_test), truth = "y_test", estimate = "null_pred")[1,]$.estimate
  # bias of CV model
  bias_optimal <- mean(y_test - optimal_pred)
  # bias of global model
  bias_null <-  mean(y_test - null_pred )
  
  # put output into a list
  tmpList <- list("testRegion" = i_eco,
    "modelObject" = fit_i,
       "modelPredictions" = tmp, 
    "performanceMetrics" = data.frame("RMSE_cvModel" = RMSE_optimal, 
                                      "RMSE_globalModel" = RMSE_null, 
                                      "bias_cvModel" = bias_optimal, 
                                      "bias_globalModel" = bias_null))

  # save model outputs
  outList[[which(sort(unique(modDat_1_s$NA_L2NAME)) == i_eco)]] <- tmpList
 }
}

## [1] "ATLANTIC HIGHLANDS"
## [1] "CENTRAL USA PLAINS"
## [1] "EVERGLADES"
## [1] "MARINE WEST COAST FOREST"
## [1] "MISSISSIPPI ALLUVIAL AND SOUTHEAST USA COASTAL PLAINS"
## [1] "MIXED WOOD PLAINS"
## [1] "MIXED WOOD SHIELD"
## [1] "OZARK/OUACHITA-APPALACHIAN FORESTS"
## [1] "SOUTHEASTERN USA PLAINS"
## [1] "UPPER GILA MOUNTAINS"
## [1] "WESTERN CORDILLERA"

Below are the RMSE and bias values for predictions made for each holdout level II ecoregion, compared to predictions from the global model for that same ecoregion

if (length(prednames_fig_1se) == 0) {
  print("The model only contains one predictor (an intercept), so cross validation isn't practical")
} else {
# table of model performance
map(outList, .f = function(x) {
  cbind(data.frame("holdout region" = x$testRegion),  x$performanceMetrics)
}
) %>% 
  purrr::list_rbind() %>% 
  kable(col.names = c("Held-out ecoregion", "RMSE of CV model", "RMSE of global model", 
                      "bias of CV model - mean(obs-pred.)", "bias of global model - mean(obs-pred.)"), 
        caption = "Performance of Cross Validation using '1 SE lambda' model specification") %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed")) 
}

Performance of Cross Validation using ‘1 SE lambda’ model specification

Held-out ecoregion	RMSE of CV model	RMSE of global model	bias of CV model - mean(obs-pred.)	bias of global model - mean(obs-pred.)
ATLANTIC HIGHLANDS	12.110219	12.024164	-4.0746600	-3.8061103
CENTRAL USA PLAINS	14.305460	14.305397	-0.2540518	-0.2504585
EVERGLADES	5.870614	5.867276	-1.8897794	-1.8796723
MARINE WEST COAST FOREST	40.621332	30.604287	-6.2761177	0.2318629
MISSISSIPPI ALLUVIAL AND SOUTHEAST USA COASTAL PLAINS	13.873606	13.878501	-0.4408670	-0.6738772
MIXED WOOD PLAINS	12.494162	12.471241	-1.9636363	-1.8159011
MIXED WOOD SHIELD	14.909675	14.841451	2.1672543	1.6848156
OZARK/OUACHITA-APPALACHIAN FORESTS	14.023692	13.906056	-3.6857508	-3.2106382
SOUTHEASTERN USA PLAINS	12.951058	12.545590	-5.7032373	-4.7645730
UPPER GILA MOUNTAINS	11.636538	11.620955	1.6233620	1.5073592
WESTERN CORDILLERA	19.296778	18.109293	6.1709829	2.2938517

if (length(prednames_fig_1se) == 0) {
  print("The model only contains one predictor (an intercept), so cross validation isn't practical")
} else {
for (i in 1:length(unique(modDat_1_s$NA_L2NAME))) {
  holdoutRegion <- outList[[i]]$testRegion
  predictionData <- outList[[i]]$modelPredictions
  modTerms <- as.matrix(coef(outList[[i]]$modelObject)) %>%
    as.data.frame() %>%
    filter(V1!=0) %>%
    rownames()

  # calculate residuals
  predictionData <- predictionData %>%
  mutate(resid = .[["obs"]] - .[["pred_opt"]] ,
         resid_globMod = .[["obs"]]  - .[["pred_null"]])


# rasterize
# use 'test_rast' from earlier

  # rasterize data
plotObs <- predictionData %>%
         drop_na(paste(response)) %>%
  #slice_sample(n = 5e4) %>%
  terra::vect(geom = c("Long", "Lat")) %>%
  terra::set.crs(crs(test_rast)) %>%
  terra::rasterize(y = test_rast,
                   field = "resid",
                   fun = mean) #%>%
  #terra::aggregate(fact = 2, fun = mean, na.rm = TRUE) %>%
  #terra::crop(ext(-1950000, 1000000, -1800000, 1000000))

tempExt <- crds(plotObs, na.rm = TRUE)

plotObs_2 <- plotObs %>% 
  crop(ext(min(tempExt[,1]), max(tempExt[,1]),
           min(tempExt[,2]), max(tempExt[,2])) 
       )

# identify locations where residuals are >100 or < -100
badResids_high <- predictionData %>% 
  filter(resid > 100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 
badResids_low <- predictionData %>% 
  filter(resid < -100)  %>% 
  terra::vect(geom = c("Long", "Lat")) %>% 
  terra::set.crs(crs(test_rast)) 


# make figures
# make histogram
hist_i <- ggplot(predictionData) +
  geom_histogram(aes(resid_globMod), col = "darkgrey", fill = "lightgrey") +
  xlab(c("Residuals (obs. - pred.)"))
# make map
map_i <-  ggplot() +
geom_spatraster(data = plotObs_2) +
  geom_sf(data = mapRegions, fill = NA, col = "rosybrown4", lwd = .5) +
  geom_sf(data=cropped_states %>% st_transform(crs = st_crs(test_rast)) %>% st_crop(st_bbox(plotObs_2)),fill=NA ) +
  geom_sf(data = badResids_high, col = "blue") +
  geom_sf(data = badResids_low, col = "red") +
labs(title = paste0("Residuals (obs. - pred.) for predictions of \n", holdoutRegion, " \n from a model fit to other ecoregions"),
     subtitle = paste0(response, " ~ ", paste0( modTerms, collapse = " + "))) +
  scale_fill_gradient2(low = "red",
                       mid = "white" ,
                       high = "blue" ,
                       midpoint = 0,   na.value = "lightgrey",
                       limits = c(-100, 100))  + 
  xlim(st_bbox(plotObs_2)[c(1,3)]) + 
  ylim(st_bbox(plotObs_2)[c(2,4)])

 assign(paste0("residPlot_",holdoutRegion),
   value = ggarrange(map_i, hist_i, heights = c(3,1), ncol = 1)
)

}

  lapply(unique(modDat_1_s$NA_L2NAME), FUN = function(x) {
    get(paste0("residPlot_", x))
  })
}

## [[1]]

## 
## [[2]]

## 
## [[3]]

## 
## [[4]]

## 
## [[5]]

## 
## [[6]]

## 
## [[7]]

## 
## [[8]]

## 
## [[9]]

## 
## [[10]]

## 
## [[11]]

Save output

# # glm models
# #mods2save <- butcher::butcher(mod_glmFinal) # removes some model components so the saved object isn't huge
# 
# #mods2save$formula <- best_form
# #mods2save$pred_vars_inter <- pred_vars_inter # so have interactions
# #n <- nrow(df_sample)
# #mods2save$data_rows <- n
# 
# 
# if(!test_run) {
#   saveRDS(mods2save, 
#         paste0("./models/glm_beta_model_CONUSwide_", s, "_n", n, 
#         #sample_group, 
#         ".RDS"))
#   if (byRegion == TRUE) {
#     ## western forests
#      saveRDS(mods2save_WF, 
#         paste0("./models/glm_beta_model_WesternForests_", s, "_n", n, 
#         #sample_group, 
#         ".RDS"))
#     ## eastern forests
#      saveRDS(mods2save_EF, 
#         paste0("./models/glm_beta_model_EasternForests_", s, "_n", n, 
#         #sample_group, 
#         ".RDS"))
#      ## grass/shrub
#      saveRDS(mods2save_G, 
#         paste0("./models/glm_beta_model_GrassShrub_", s, "_n", n, 
#         #sample_group, 
#         ".RDS"))
#   }
# }

## partial dependence plots
#vip::vip(mod_glmFinal, num_features = 15)

#pdp_all_vars(mod_glmFinal, mod_vars = pred_vars, ylab = 'probability',train = df_small)

#caret::varImp(fit)

session info

Hash of current commit (i.e. to ID the version of the code used)

system("git rev-parse HEAD", intern=TRUE)

## [1] "79890c55a196d40eb16ae968701c4515b44c260c"

Packages etc.

sessionInfo()

## R version 4.4.0 (2024-04-24)
## Platform: aarch64-apple-darwin20
## Running under: macOS Sonoma 14.7.5
## 
## Matrix products: default
## BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib 
## LAPACK: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.12.0
## 
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
## 
## time zone: America/Denver
## tzcode source: internal
## 
## attached base packages:
## [1] parallel  stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] doMC_1.3.8                 iterators_1.0.14           foreach_1.5.2              ggpubr_0.6.0               factoextra_1.0.7          
##  [6] USA.state.boundaries_1.0.1 glmnet_4.1-8               Matrix_1.7-0               kableExtra_1.4.0           rsample_1.2.1             
## [11] here_1.0.1                 StepBeta_2.1.0             ggtext_0.1.2               knitr_1.49                 gridExtra_2.3             
## [16] pdp_0.8.2                  GGally_2.2.1               lubridate_1.9.4            forcats_1.0.0              stringr_1.5.1             
## [21] dplyr_1.1.4                purrr_1.0.4                readr_2.1.5                tidyr_1.3.1                tibble_3.2.1              
## [26] tidyverse_2.0.0            caret_6.0-94               lattice_0.22-6             ggplot2_3.5.1              sf_1.0-20                 
## [31] tidyterra_0.6.1            terra_1.8-21               ggspatial_1.1.9            dtplyr_1.3.1               patchwork_1.3.0           
## 
## loaded via a namespace (and not attached):
##   [1] RColorBrewer_1.1-3   rstudioapi_0.17.1    jsonlite_1.9.1       shape_1.4.6.1        magrittr_2.0.3       modeltools_0.2-23   
##   [7] farver_2.1.2         rmarkdown_2.29       vctrs_0.6.5          rstatix_0.7.2        htmltools_0.5.8.1    broom_1.0.7         
##  [13] Formula_1.2-5        pROC_1.18.5          sass_0.4.9           parallelly_1.37.1    KernSmooth_2.23-22   bslib_0.9.0         
##  [19] plyr_1.8.9           sandwich_3.1-0       zoo_1.8-12           cachem_1.1.0         commonmark_1.9.1     lifecycle_1.0.4     
##  [25] pkgconfig_2.0.3      R6_2.6.1             fastmap_1.2.0        future_1.33.2        digest_0.6.37        colorspace_2.1-1    
##  [31] furrr_0.3.1          rprojroot_2.0.4      pkgload_1.3.4        labeling_0.4.3       yardstick_1.3.1      timechange_0.3.0    
##  [37] mgcv_1.9-1           abind_1.4-8          compiler_4.4.0       proxy_0.4-27         aod_1.3.3            withr_3.0.2         
##  [43] backports_1.5.0      carData_3.0-5        betareg_3.1-4        DBI_1.2.3            ggstats_0.9.0        ggsignif_0.6.4      
##  [49] MASS_7.3-60.2        lava_1.8.0           classInt_0.4-10      gtools_3.9.5         ModelMetrics_1.2.2.2 tools_4.4.0         
##  [55] units_0.8-5          lmtest_0.9-40        future.apply_1.11.2  nnet_7.3-19          glue_1.8.0           nlme_3.1-164        
##  [61] gridtext_0.1.5       grid_4.4.0           reshape2_1.4.4       generics_0.1.3       recipes_1.1.0        gtable_0.3.6        
##  [67] tzdb_0.4.0           class_7.3-22         data.table_1.17.0    hms_1.1.3            utf8_1.2.4           car_3.1-2           
##  [73] xml2_1.3.7           flexmix_2.3-19       markdown_1.13        ggrepel_0.9.5        pillar_1.10.1        splines_4.4.0       
##  [79] survival_3.5-8       tidyselect_1.2.1     svglite_2.1.3        stats4_4.4.0         xfun_0.51            hardhat_1.4.0       
##  [85] timeDate_4032.109    stringi_1.8.4        yaml_2.3.10          evaluate_1.0.3       codetools_0.2-20     cli_3.6.4           
##  [91] rpart_4.1.23         systemfonts_1.2.1    munsell_0.5.1        jquerylib_0.1.4      Rcpp_1.0.14          globals_0.16.3      
##  [97] gower_1.0.1          listenv_0.9.1        viridisLite_0.4.2    ipred_0.9-15         scales_1.3.0         prodlim_2024.06.25  
## [103] e1071_1.7-14         crayon_1.5.3         combinat_0.0-8       rlang_1.1.5          cowplot_1.1.3